amp-bind: Binding limits vs. infinite scroll

[dreamofabear]

Context: #11434 (comment)

amp-bind's bindings limit will be problematic once infinite scroll or appending "view more" use cases become more prevalent.

One of the original reasons why these limits exist is (background) initialization latency for amp-bind. If parsing an expression takes O(1ms) on modest hardware, then 2000 bindings means the AMP page could have up to 2s time-to-interactive (TTI) WRT amp-bind.

For infinite scroll/view more, the incremental markup added per trigger is typically small. So we likely need not worry about TTI. Instead, we should think about the total evaluation time for updating the page with a larger number of bindings when AMP.setState() is invoked. Luckily, evaluating expressions are typically much faster than parsing them [1], so perhaps something like the following could work:

• Rework 1000 binding limit to only apply to initial DOM scan
• Allow X >> 1000 total binding limit -- includes runtime changes e.g. infinite scroll/view more
• Allow Y << 1000 incremental binding limit -- which constrains the number of bindings that may be added per infinite scroll/view more trigger [2]

[1] One issue is that the expression complexity and binding limits are only coarse approximations for amp-bind runtime performance. For example, we don't factor in the asymptotic complexity for functions like concat() which are highly dependent on the size of input data.

[2] Need to consider the fact that expression parse/evaluation is render-blocking by default in amp-list. Parse cost can be mitigated with use of amp-bind-macro since the expressions are likely very similar. Definitely feeling more mixed about that change in hindsight.

/cc @ericlindley-g @aghassemi

[aghassemi]

Great write up. Few thoughts:
1- Can incremental and initial limits be the exact same 1000 or are you suggesting incremental be like 500? I expect initial >> incremental.
2- Is amp-bind limits per-doc or per-window? Thinking about the content-level infinite scroll which currently can load 2 more full amp docs. Ideally the newly loaded AMP doc is treated as a separate context, specially given reevaluation is limited to its scope.
3- Total binding makes sense to me as a formula of n * incremental + initial where n is some reasonable expected load more hits. 5? so based on numbers above 3500, is that reasonable?

[dreamofabear]

1. Yea, I was thinking ~100-250 incremental limit. I think users have a lower tolerance for interactivity latency after page load vs. on page load, so it should be significantly lower than initial.
2. The Bind service is scoped per-doc and scans the document body, so it should work as expected.
3. Exactly, I was thinking ~2000-3000.

[dreamofabear]

Similar request: #17233

[ampprojectbot]

This issue hasn't been updated in awhile. @choumx Do you have any updates?

[ampprojectbot]

This issue hasn't been updated in awhile. @choumx Do you have any updates?

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed in 7 days if no further activity occurs. Thank you for your contributions.